Thank you Unbird for your help.. And thank you for your time testing this with your engine !
I'm sorry, I didn't understand posts from the link you gave me...
Ok, it's helpfull to know that edgeX is opposite to vertexX. But what else ?
You said in the other post :
As long as neighboring patches share the same control points, and the control point to tessellation factor algorithm will produce the same output for the given input control points, then the order doesn't make any difference.
But as the picture I posted before (ico from upside), I cannot produce the same control points without altering the order, which will break the cull back face. Or I'm missing something ?
I thought about using a 4 points patch list, the fourth point is the order to give to the edges.
It almost works, but creates strange artifacts.. Perhaps the bug you said about the June 2010 SDK (I'm using it) ?